Although traditional software reliability growth models, SRGMs, have been very intensely studied and investigated, there are still many problems that have been responsible for their possible inaccurate predictions in practice. Here, some of these problems and proposed solutions are discussed. We use two enhancement approaches (recalibration and model combination) in order to try to improve the accuracy of predictions when using this conventional type of modelling. These two approaches share the idea of reuse of the original prediction systems, which are obtained from analyzing test data in a more productive way. Numerical examples are presented to investigate actual software failure data are conducted in order to investigate the effectiveness of the two approaches. In these examples we have used two models, Weibull and Burr Type XII with different choices of parameters alongside a base line model. Further, we have used Prequential Likelihood Ratio, u-plot, and y-plot to examine the predictive performance of the underlying models before and after applying the two development approaches. We have shown that if used in isolation the refinement approaches may still yield poor prediction results. However, if model combination is used on the recalibrated models, then stable prediction results can be achieved even for the data that has been collected under less carefully controlled conditions than in typically the case for SRGMs.